SAS 9.4 Language Reference, 6th Edition

patterns of instruction and replaces them with more efficient sequences, and also

performs optimizations that pertain to the SAS register. In most cases, performing the

code-generation optimization is preferable. If you have a large DATA step program,

performing code generation optimization can result in a significant increase in

compilation time and overall execution time.

You can reduce or turn off the code generation optimization by using the

CGOPTIMIZE= system option. Set the code generation optimization that you want SAS

to perform using these CGOPTIMIZE= system option values:

0 performs no optimization during code compilation.

1 specifies to perform stage 1 optimization. Stage 1 optimization removes

redundant instructions, missing value checks, and repetitive computations for

array subscripts; detects patterns of instructions and replaces them with more

efficient sequences.

2 specifies to perform stage 2 optimization. Stage 2 performs optimizations that

pertain to the SAS register. Performing stage 2 optimization on large DATA step

programs can result in a significant increase in compilation time.

3 specifies to perform full optimization, which is a combination of stages 1 and 2.

This is the default value.

For more information, see “CGOPTIMIZE= System Option” in SAS System Options:

Reference.

Calculating Data Set Size

If you have already applied optimization techniques but still experience lengthy

processing times or excessive memory usage, the size of your data sets might be very

large. In that case, further improvement might not be possible.

You can estimate the size of a data set by creating a dummy data set that contains the

same variables as your data set. Run the CONTENTS procedure, which shows the size

of each observation. Multiply the size by the number of observations in your data set to

obtain the total number of bytes that must be processed. You can compare processing

statistics with smaller data sets to determine whether the performance of the large data

sets is in proportion to their size. If not, further optimization might still be possible.

Note: When you use this technique to calculate the size of a data set, you obtain only an

estimate. Internal requirements, such as the storage of variable names, might cause

the actual data set size to be slightly different.

206 Chapter 12 • Optimizing System Performance